ComfyUI Mochi Edit Workflow Text to Video

These nodes are built to work with the ComfyUI-MochiWrapper nodes and soon will work with native ComfyUI Mochi too. For now please follow the installation for the wrapper.

Then git clone this repo into your ComfyUI/custom_nodes/ directory or use the ComfyUI Manager to install (when this repo is added there).

There are no additional requirements.

This node creates a sampler that can convert the noise into a video.

• latents: the latents of the original video
• eta: the strength that the generation should align with the original video
  • higher values lead the generation closer to the original
• start_step: the starting step to where the original video should guide the generation
  • a lower value (e.g. 0) will have much closer following but not allow for additional objects like a hat to be placed
  • a higher value (e.g. 6) will allow for new objects like a hat to be placed, but may not follow the original video. Higher values can also lead to bad results (blurs)
• end_step the step to stop guiding the generation closer to the original video
  • a lower value will lead to more differences in the video output
• eta_trend: whether the eta (strength of guidance) should stay constant, increase, or decrease as steps progress. linear_decrease is the recommended setting for most changes.

This is the classic KSampler or SamplerCustom from ComfyUI but for the MochiWrapper.

• positive and negative can be anything you like. positive shoud be the target prompt.
• cfg: can have any cfg that would work with normal Mochi (e.g. 4.50)
• latents: should be the latent from unsampling
• sigmas: must be prepared but NOT flipped
• seed: the seed has no effect